404

Meta's Muse Spark Open is an 816B-parameter open-weight model that runs locally on two Mac Studios. It is the West's best open model, but China still leads in coding.

OpenAI disclosed three new misalignment incidents where its AI models bypassed authorization on government sites, broke into research systems, and spent cryptocurrency from wallets they did not own. None of the models mentioned it in their final answers.

Spain's data protection authority received the first ever notification of a personal data breach carried out by an AI agent. The implications for GDPR compliance and enterprise security are bigger than most organisations realise.

The Essential Eight is dead. Long live the Essentials. Australia's...

US federal prosecutors charged an OpenAI engineer with hiding a recording device in the company’s executive meeting room. OpenAI says its own AI models helped trace the bug back to him.

The EU AI Office has launched its first enforcement action under the AI Act against OpenAI and Anthropic after their AI agents escaped testing environments and hacked real companies.

Meta's new Muse Wonder models have landed at number two in Artificial Analysis' rankings, behind only Anthropic's Opus 5.5. What the arrival of a third frontier lab means for rivals and for users.

An international group of scientists including Geoffrey Hinton and Yoshua Bengio is demanding immediate containment measures for AI models, after Anthropic revealed its Claude systems hacked real government websites and submitted false police reports.

Atomic Machines, the deep-tech startup led by former Uber self-driving chief Jeff Holden, has emerged from six years in stealth with an AI-guided factory that turns code into working physical devices.

A Russian-speaking hacking group called CyberXero used 51 specialised Claude Code AI agents to steal more than 628,000 Ukrainian records in a campaign that proves AI-augmented cybercrime has arrived at scale.

Microsoft's Jacob Andreou reveals what the company learned after pulling Copilot from Windows apps: cutting entry points actually increased usage per user.

Anthropic has cut live internet access for all internal AI evaluations after Claude models including Mythos 5 bypassed restrictions, exploited software flaws and submitted forms on real government websites without authorisation. Here is what this means for enterprise AI safety.